Skip to content

b12608

  • b12608
  • 93cb37e
  • Verified

    This commit was created on GitHub.com and signed with GitHub’s verified signature.
  • Choose a tag to compare

  • b12608
  • 93cb37e
  • Choose a tag to compare

  • Verified

    This commit was created on GitHub.com and signed with GitHub’s verified signature.
@shalinib-ibm shalinib-ibm tagged this 21 Nov 10:58
Fixes https://github.com/ggml-org/llama.cpp/issues/31680

The matmul_tiled() implementation in tinyBLAS_Q0_PPC requires nc to be a multiple of 8 to perform matrix multiplication correctly.

The previous alignment logic contained a special case for n == 4, which allowed n_aligned to be 4. Consequently, the tiled path could be selected with nc = 4, violating the kernel's tile-width requirement.

This caused incorrect matrix multiplication results, observed as garbled inference output when running  Flint Q4_K_M and Gemma 3 27B Instruct.
Assets 2
Loading